Meta's Muse AI agent surges to top app charts amid security fixes

Big Tech's rush to dominate artificial intelligence has exposed a pattern of corporate overreach, from the quiet appropriation of vast troves of creative work to the hasty rollout of tools riddled with security holes. Court documents unsealed in a lawsuit filed by Mother Jones and other publishers against OpenAI and Microsoft reveal executives privately acknowledging what critics have long argued: the industry's foundation rests on the unauthorized use of copyrighted material at an unprecedented scale.

Microsoft's director of applied science, Brent Hecht, described a emerging "doom loop" in which AI systems erode the economic base of the very suppliers—newsrooms, writers, and researchers—whose output trained the models. The documents, filed September 4 in ongoing litigation, quote company insiders framing the training process as potentially "the largest theft of labor in human history." Similar suits from The New York Times and The Intercept underscore the same grievance: AI developers ingested millions of articles and books without permission or compensation, then packaged the results as premium products.

This acknowledgment clashes sharply with the industry's public messaging. Meta's chief AI officer, Alexandr Wang, has spent recent weeks promoting the company's new Muse app as a financial lifesaver for ordinary users. The tool, launched September 8, is pitched as capable of reviewing bank statements, canceling unused subscriptions, and haggling with insurers to claw back cash—prompting Wang to launch the #MuseMoneyChallenge on social media. Early adopters reported savings in the thousands, including one user who recovered a year's worth of forgotten book-subscription fees. Wang has framed the agent as an instant path to "$1,000" in reclaimed money.

Yet the same week brought confirmation that Muse itself carried a serious vulnerability. Security researcher Patrick Wardle identified a zero-day flaw in the macOS version that allowed attackers with local device access to redirect the app's cloud transcription to their own servers. The exploit could grant control over the AI agent, enabling it to capture screenshots or write malicious files without alerting the user. Meta issued a patch within hours of the report but maintained that real-world risk was low because the attack required physical proximity. Critics noted the irony: a product sold on convenience and privacy was built with design choices—cloud processing and undocumented settings—that left users exposed.

The contradiction is difficult to ignore. Companies racing to monetize AI simultaneously undermine the creators whose labor powers it and release consumer-facing products before basic safeguards are in place. Hecht's internal warning about a "doom loop" now reads as a broader forecast. If news outlets and independent writers can no longer sustain themselves because their work has been absorbed without payment, the supply of reliable information feeding future models will shrink. At the same time, users are encouraged to hand over financial details to agents that proved susceptible to hijacking.

Lawsuits and security disclosures are forcing a reckoning that marketing campaigns have tried to sidestep. The scale of the data extraction documented in the Mother Jones case, paired with the rapid discovery of exploitable flaws in tools like Muse, suggests the industry's growth model depends on externalizing costs—onto creators, onto users, and ultimately onto the information ecosystem itself. Regulators and courts will decide how much longer that model can continue.